Unlock instant, AI-driven research and patent intelligence for your innovation.

Training method for object weight recognition system, object weight recognition method and related equipment

A training method and re-recognition technology, applied in character and pattern recognition, instrumentation, computing, etc., can solve problems such as poor accuracy, poor performance of feature extractors, and feature extractors affecting the accuracy of object re-recognition, and achieve great flexibility , the effect of slowing down overfitting and improving performance

Active Publication Date: 2021-01-08
BEIJING TUSEN ZHITU TECH CO LTD
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Therefore, the performance of the feature extractor directly affects the accuracy of object weight recognition. At present, there are technical problems of poor performance of feature extractors, and there are technical problems of poor accuracy of object weight recognition using this feature extractor.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Training method for object weight recognition system, object weight recognition method and related equipment
  • Training method for object weight recognition system, object weight recognition method and related equipment
  • Training method for object weight recognition system, object weight recognition method and related equipment

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0058] Embodiment 1 of the present invention provides a training method for an object re-identification system. After the feature extractor of the object re-identification system, a random walk module and a classifier are sequentially arranged, such as figure 1 As shown, the output of the feature extractor is the input of the random walk module, and the product of the output of the feature extractor and the output of the random walk module is the input of the classifier.

[0059] see figure 2 , is a flow chart of the training method of the object re-identification system provided by the embodiment of the present invention. According to multiple sets of training sample images, the object re-identification system is trained in the following order. For the convenience of description, the embodiment of the present invention will process each The set of training sample images is called the current set of training sample images:

[0060] Step 101. Input the current group of traini...

Embodiment 2

[0093] Embodiment 2 of the present invention provides a method for object re-identification, in which a random walk model is set after the feature extraction network, such as Figure 7 shown. The flow of the object re-identification method is as follows: Figure 8 shown, including:

[0094] Step 201, input the image to be searched and the gallery image in the gallery into the feature extraction network to obtain the feature vectors of the image to be searched and the gallery image;

[0095] Step 202. For each image to be searched, perform the following steps 202A to 202E to obtain the image sequence corresponding to each image to be searched:

[0096] Step 202A, calculating the similarity between the image to be searched and the features of each gallery image, and arranging each gallery image in order of similarity from high to low to obtain an image sequence;

[0097] Step 202B, selecting the first M images from the image sequence, and the feature vectors of the M images con...

Embodiment 3

[0112] Based on the same idea as the previous embodiment 1, embodiment 3 of the present invention provides a method for object weight identification, the flow of the method is as follows Figure 10 shown, including:

[0113] Step 301, input the image to be searched and the gallery image in the gallery into the feature extraction network to obtain the feature vectors of each image to be searched and gallery image; the feature extraction network is any object weight provided in the first embodiment in advance. The training method of the recognition system is trained (for details, please refer to the relevant content in Embodiment 1, and will not repeat them here);

[0114] Step 302. For each image to be searched, perform the following steps 302A to 302B to obtain the image sequence corresponding to each image to be searched:

[0115] Step 302A, calculating the similarity between the image to be searched and the features of each gallery image, and arranging each gallery image in...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a training method for an object re-identification system, an object re-identification method and related equipment, so as to improve the performance of a feature extractor, thereby improving the accuracy of object re-identification using the feature extractor. The method includes: inputting the current group of training sample images into a feature extractor to obtain feature vectors of each training sample image, and the feature vectors of each training sample image form a feature vector group; each feature vector in the feature vector group is used as A node processes the eigenvector group through a random walk module to obtain a similarity matrix representing the similarity between each eigenvector, and inputs the product of the eigenvector group and the similarity matrix into a classifier , obtain the classification result of classifying the current group of training sample images; adjust the parameters of the feature extractor and classifier according to the classification results, and process the next group of training sample images based on the adjusted feature extractor and classifier.

Description

technical field [0001] The present invention relates to the field of computer deep learning, in particular to a training method and device for an object re-identification system, an object re-identification method and device, and a processing device. Background technique [0002] Object weight recognition technology plays an essential role in the fields of intelligent video surveillance, robotics, and automatic driving. Given an image of an object to be retrieved, object re-identification technology aims to retrieve related images identical to the object from images captured by different cameras. The influence of camera angle, object pose, occlusion, etc. makes the object re-identification task quite challenging. [0003] Thanks to the emergence of deep learning technology, object weight recognition technology has developed rapidly in recent years. Most of the current state-of-the-art methods are based on deep learning and mainly consist of two parts, feature extractor and...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06K9/62G06F16/583
CPCG06F18/22G06F18/214G06F18/24
Inventor 罗传琛陈韫韬王乃岩
Owner BEIJING TUSEN ZHITU TECH CO LTD